Twin Prime No

712 RESONANCE August 2013
GENERAL ARTICLE
The Twin Prime Problem and Generalizations
(aprs Yitang Zhang)
M Ram Murty
Keywords
Twin primes, Bruns theorem,
von Mangoldt function, sieves in
number theory, Bombi eri
Vinogradov method, pirmes in
arithmetic progressions.
M Ram Murty, a Canadian
mathematician, is
currently Queen's
Research Chair in
Mathematics and
Philosophy at the Queen's
University in Canada. He
is also an adjunct
professor at several
institutions in India (TIFR
in Mumbai, IMSc in
Chennai and HRI in
Allahabad.) He also
teaches Indian philosophy
at Queen's University and
has recently published a
book titled Indian
Philosophy and has written
many popular monographs
for students of
mathematics.
We give a short introduction to the recent break-
through theorem of Yitang Zhang that there are
innitely many pairs of distinct primes (p, q) with
[p q[ < 70 million.
The twin prime problem asks if there are innitely many
primes p such that p + 2 is also prime. More gener-
ally, one can ask if for any even number a, there are
innitely many primes p such that p + a is also prime.
This problem inspired the development of modern sieve
theory. Though several sophisticated tools were discov-
ered, the problem deed many attempts to resolve it
until recently.
On April 17, 2013, a relatively unknown mathematician
from the University of New Hampshire, Yitang Zhang,
submitted a paper to the Annals of Mathematics. The
paper claimed to prove that there are innitely many
pairs of distinct primes (p, q) with [p q[ < 7 10
7
.
This was a major step towards the celebrated twin prime
conjecture! A quick glance at the paper convinced the
editors that this was not a submission from a crank. The
paper was crystal clear and demonstrated a consummate
understanding of the latest technical results in analytic
number theory. Therefore, the editors promptly sent it
to several experts for refereeing. The paper was accepted
three weeks later.
In this article, we will outline the proof of this recent
breakthrough theorem of Yitang Zhang [1]. Even though
this article is only an outline, it should help the seri-
ous student to study Zhangs paper in greater detail.
An essential ingredient in Zhangs proof is the idea of
smoothness which allows him to extend the range of
GENERAL ARTICLE
applicability of earlier theorems. (A number is said to
be y-smooth if all its prime factors are less than y.) The
rudimentary background in analytic number theory is
readily obtained from [2] and [3]. This can be followed
by a careful study of [4] and the three papers [5], [6], [7].
1. Introduction and History
Let p
1
, p
2
, ... be the ascending sequence of prime num-
bers. The twin prime problem is the question of whether
there are innitely many pairs of primes (p, q) with
[p q[ = 2. This problem is usually attributed to the
ancient Greeks, but this is very much Greek mythol-
ogy and there is no documentary evidence to support it.
The rst published reference to this question appeared
in 1849 by Alphonse de Polignac who conjectured more
generally that for any given even number 2a, there are
innitely many pairs of primes such that [p q[ = 2a.
Box 1.
In the 1912 International Congress of Mathematicians in Cambridge, England, the renowned mathematician
Edmund Landau cited the following four problems as hopeless (in terms of finding a solution in the near
future):
1. Every even number greater than two is the sum of two primes.
2. Between any consecutive squares there is a prime number.
3. There are infinitely many twin primes.
4. There are infinitely many primes which are one more than a perfect square.
Even today, the status of each of the problems except the third one cannot be considered completely different.
The third problem on twin primes is said to be from the ancient Greek times, but the first published reference
is due to Alphonse de Polignac in 1849.
The problem asks whether there are infinitely many pairs p, p+2 which are both primes. The number of such
pairs even if shown to be infinite is rather sparse, in the sense that the sum of their reciprocals is a finite
number. No finite number N was known to exist having the property that there are infinitely many pairs p,
p + N both of which are primes.
On the 17th of April, 2013, a comparatively unknown mathematician named Yitang Zhang, submitted a proof
that there are infinitely many pairs p,q of primes which differ by less than 7 crores. This is a major breakthrough
vis-a-vis the twin prime problem!
B Sury, Editor
GENERAL ARTICLE
Zhangs proof
depends on major
milestones of 20th
century number
theory and algebraic
geometry. Thus, it is
definitely a 21st
century theorem!
In a recent paper [1] in the Annals of Mathematics, Yi-
tang Zhang proved that there are innitely many pairs
of distinct primes (p, q) with
[p q[ < 7 10
7
.
His proof depends on major milestones of 20th century
number theory and algebraic geometry. Thus, it is def-
initely a 21st century theorem! Undoubtedly, his paper
opens the door for further improvements and it is our
goal to discuss some of these below.
After de Polignacs conjecture, the rst serious paper
on the subject was by Viggo Brun in 1915, who, after
studying the Eratosthenes sieve, developed a new sieve,
now called the Brun sieve, to study twin primes and
related questions. He proved that
p:p+2 prime
1
p
< .
By contrast, the sum of the reciprocals of the primes
diverges and so, this result shows that (in some sense)
if there are innitely many twin primes, they are very
sparse.
A few years later, in 1923, Hardy and Littlewood [8],
made a more precise conjecture on the number of twin
primes up to x. They predicted that this number is (see
p.371 of [9])
2
p>2
1
1
(p 1)
2
x
log
2
x
.
Here, the symbol A(x) B(x) means that A(x)/B(x)
tends to 1 as x tends to innity.
They used the circle method, originally discovered by
Ramanujan and later developed by Hardy and Ramanu-
jan in their research related to the partition function.
After Ramanujans untimely death, it was taken further
GENERAL ARTICLE
The circle method
has the potential to
make precise
conjectures
regarding additive
questions such as
the Goldbach
conjecture and the
twin prime problem.
by Hardy and Littlewood in their series of papers on
Warings problem. (The circle method is also called the
HardyLittlewood method by some mathematicians.)
In the third paper of this series, they realized the po-
tential of the circle method to make precise conjectures
regarding additive questions, such as the Goldbach con-
jecture and the twin prime problem.
Based on heuristic reasoning, it is not dicult to see
why such a conjecture should be true. The prime num-
ber theorem tells us that the number of primes (x), up
to x, is asymptotically x/ log x. Thus, the probability
that a random number in [1, x] is prime is 1/ log x and so
the probability that both n and n+2 are prime is about
1/ log
2
x. The constant is a bit more delicate to conjec-
ture and is best derived using the theory of Ramanujan
Fourier series expansion of the von Mangoldt function
as in a recent paper of Gadiyar and Padma [8] (see also
[11] for a nice exposition). However, it is possible to
proceed as follows. By the unique factorization theorem
of the natural numbers, we can write
log n =
d|n
(d),
where (d) = log p if d is a power of a prime p and zero
otherwise. This is called the von Mangoldt function. By
the Mobius inversion formula, we have for n > 1
(n) =
d|n
(d) log
n
d
=
d|n
(d) log d,
since
d|n
(d) = 1 if n = 1 and zero otherwise. Thus,
to count twin primes, it is natural to study
px
(p + 2),
where the sum is over primes p less than x. Using the
formula for (n), the sum above becomes
px
d|p+2
(d) log d =
dx+h
(d) log d
px,p2(mod d)
1.
GENERAL ARTICLE
The innermost sum is the number of primes p x that
are congruent to 2 (mod d), which for d odd is asymp-
totic to (x)/(d), where (x) is the number of primes
upto x, and is Eulers function. Ignoring the error
terms, our main term now is asymptotic to
(x)
dx+2,dodd
(d) log d
(d)
.
This suggests that
px
(p + 2) (x)
d>1,d odd
(d) log d
(d)
.
The innite series in the brackets is not absolutely con-
vergent. However, it converges conditionally and can be
evaluated as follows.
Consider the Dirichlet series
F(s) :=
d=1
(d)
(d)d
s
,
which converges absolutely for '(s) > 0. Now, F(s)
admits an Euler product
p,(p,2)=1
1
1
p
s
(p 1)
which resembles 1/(s + 1) (with the 2-Euler factor re-

moved) and so it is natural to write the product as
(s+1)
1
(12
s1
)
1
p>2
1
1
p
s+1
1
1
p
s
(p 1)
.
It is now easy to see that the Euler product converges
absolutely for '(s) 0 and this gives an analytic con-
tinuation of F(s) for '(s) 0. The twin prime constant
is now F
(0) and because (s + 1) has a simple pole at

GENERAL ARTICLE
s = 0 with residue 1, the term (s + 1)
1
has a zero at
s = 0. Thus, our heuristic reasoning gives
px
(p + 2) 2(x)
p>2
1
1
(p 1)
2
,
which agrees with the HardyLittlewood conjecture (af-
ter applying partial summation). A similar argument
provides the conjectured formula of Hardy and Little-
wood for the number of prime pairs that dier by an
even number 2h.
The reader will note that one can make the above argu-
ment precise by introducing the error terms
E(x, d, a) := (x, d, a)
(x)
(d)
and it is easy to see that the error term in our calculation
is
dx+2
E(x, d, 2).
The BombieriVinogradov theorem states that
dQ
[E(x, d, 2)[
x
log
A
x
for any A > 0 and Q x
1/2
log
B
x, where B = B(A)
is a function of A. In fact, one can take B(A) = A + 5
(see page 161 of [3]). Elliott and Halberstam [12] have
conjectured that the result is valid for any Q < x
1
for any > 0. Even admitting this conjecture, we see
that the interval [x
1
, x] still needs to be treated. It
is this obstacle that motivates the use of truncated von
Mangoldt functions:
D
(n) :=
d|n,d<D
(d) log(D/d)
and more generally
D
(n; a) :=
1
a!
d|n,d<D
(d) log
a
(D/d)
GENERAL ARTICLE
as will be indicated below.
2. The Basic Strategy of Zhangs Proof
Let (n) = log n if n is prime and zero otherwise. We
will use the notation n x to mean that x < n < 2x.
Now, suppose we can nd a positive real-valued function
f such that for
S
1
=
nx
f(n),
S
2
=
nx
((n) + (n + 2)) f(n) ,
we have
S
2
(log 3x)S
1
> 0,
for suciently large x. Then we can deduce that there
exists an n such that n and n + 2 are both prime with
x < n < 2x. Such a technique and a method to choose
optimal functions f goes back to the 1950s and is rooted
in the Selberg sieve. See for example [2] for a short
introduction to the Selberg sieve.
The problem as posed above is intractable. So we gen-
eralize the problem and consider sets
H = h
1
, h
2
, ..., h
k
.
It is reasonable to expect (under suitable conditions)
that there are innitely many n such that n + h
1
, n +
h
2
, ..., n+h
k
are all prime. This would be a form of the
generalized twin prime problem and was rst enunciated
in the paper by Hardy and Littlewood alluded to above.
Clearly, we need to put some conditions on H. Indeed,
if for some prime p the image of H (mod p) has size p,
then all the residue classes are represented by p so that
in the sequence,
n + h
1
, n + h
2
, ..., n + h
k
there will always be some element divisible by p and it is
unreasonable to expect that for innitely many n all of
GENERAL ARTICLE
these numbers are prime numbers. So a necessary con-
dition is that
p
(H) = [H[ (mod p) < p for every prime
p. Under such a condition, the set is called admissible
and we expect this to be the only local obstruction.
Zhang [1] proves:
Theorem 1. Suppose that H is admissible with k
3.5 10
6
. Then, there are innitely many positive inte-
gers n such that the set
n + h
1
, ..., n + h
k
contains at least two primes. Consequently,

liminf
n
(p
n+1
p
n
) < 7 10
7
.
In other words, p
n+1
p
n
is bounded by 7 10
7
for in-
nitely many n.
Zhang shows that the second assertion follows from the
rst if we choose for H a set of k
0
= 3.5 10
6
primes
lying in the interval [3.510
6
, 710
7
]. This can be done
since
(7 10
7
) (3.5 10
6
) > 3.5 10
6
from known explicit upper and lower bounds for (x)
due to Dusart [13]. That such a set of primes is admissi-
ble is easily checked. Indeed, if p > k
0
,
p
(H) k
0
< p.
If p < k
0
and
p
(H) = p, then one of the prime elements
is divisible by p and hence equal to p, a contradiction
since we chose elements of H to be primes > k
0
.
The main strategy of the proof goes back to the paper
by Goldston, Pintz and Yildrin [4] where they consider
S
1
=
nx
f(n),
S
2
=
nx

hH
(n + h)
f(n).
GENERAL ARTICLE
The main strategy
of Zhangs proof
goes back to the
paper by
Goldston, Pintz and
Yildrin.
The idea is to show that for some admissible H, we have
S
2
(log 3x)S
1
> 0.
This would imply that there are at least two primes
among the sequence
n + h
1
, ..., n + h
k
.
They choose, f(n) = (n)
2
with
(n) =
1
(k + )!
d|P(n),d<D
(d)g(d),
where denotes the familiar M obius function and
g(d) =
log
D
d
k+
,
and
P(n) =
hH
(n + h).
What is now needed is a good upper bound for S
1
and
a good lower bound for S
2
. This is the same strategy
adopted in [4]. To elaborate, let C
i
(d) be the set of
solutions (mod d) for P(n h
i
) 0 (mod d) and dene
the singular series S by
S =
1

p
(H)
p

1
1
p
k
.
With
T
1
=
1
(k + 2)!
S(log D)
k+2
+ o((log x)
k+2
)
and
T
2
=
1
(k + 2 + 1)!
2 + 2
+ 1
S(log D)
k+2+1
+ o((log x)
k+2+1
),
GENERAL ARTICLE
the argument of [4] leads to
S
2
(log 3x)S
1
= (kT
2
(log x)T
1
)x
+ O(x(log x)
k+
) + O(E) ,
where
E =
1ik
d<D
2
(d)
3
(d)
k1
(d)
cC
i
(d)
(; d, c)
and
(; d, c) =
nx,nc(mod d)
(n)
1
(d)
nx
(n).
We write a(x) = o(b(x)) if a(x)/b(x) tends to zero as x
tends to innity. We also write for non-negative b(x),
a(x) = O(b(x)) (or a(x) b(x)) if there is a constant
K such that [a(x)[ Kb(x) for all x.
Let us look at the main term. A quick calculation shows
that it is
2k(2 + 1)
(k + 2 + 1)( + 1)
log D log x
(log D)
k+2
(k + 2)!
Sx.
We need to choose D so that the term in brackets is
positive. Let D = x
. The term in brackets is positive

provided
2k(2 + 1)
(k + 2 + 1)( + 1)
1 > 0.
That is, we need
>
(k + 2 + 1)( + 1)
2k(2 + 1)
=
1
4
1 +
2 + 1
k

1 +
1
2 + 1
.
GENERAL ARTICLE
The constant k in
Zhangs theorem has
been reduced to 1466
and the gap between
consecutive primes is
now at most 12006
infinitely often.
From this, we see that if k and are chosen to be suf-
ciently large and /k is suciently small, the quantity
on the right side is asymptotic to 1/4. Thus, if we can
choose > 1/4 then we can nd choices of k and for
which the main term is positive.
The error term is easily recognized to be related to the
BombieriVinogradov theorem which shows that for any
< 1/4, the error is negligible. So we seem to be at an
impasse. However, a well-known conjecture of Elliott
and Halbertsam [12] predicts that the error is negligible
for any < 1/2. This is where things stood in 2005
after the appearance of the paper [4].
The new contribution of Zhang is that in the sums T
1
and T
2
(which are actually dened as terms involving
the Mobius function and g(d)), he notes that terms with
divisors d having a large prime divisor are relatively
small. So if we let P be the product of primes less than
a small power of x and impose the condition that d[P
in the BombieriVinogradov theorem then he is able to
establish the following:
Theorem 2. For 1 i k, we have
d<D
2
,d|P
cC
i
(d)
[(; d, c)[
x
log
A
x
,
for any A > 0 and D = x
with = 1/4 + 1/1168.

This theorem is the new innovation. Zhang admits that
his choice of k may not be optimal and that the optimal
value of k is an open problem that will not be discussed
in this paper.
After the appearance of Zhangs paper, several blogs
have discussed improvements, the most notable being
Taos blog [14] and and another blog [15], where (as of
July 10, 2013), k in Zhangs theorem has been reduced
to 1466 and the gap between consecutive primes is now
GENERAL ARTICLE
at most 12,006 innitely often. These are encouraging
developments and perhaps we are well on our way to
resolving the twin prime conjecture in the foreseeable
future.
3. A Closer Look at S
1
and S
2
The analysis of S
1
and S
2
follows closely the treatment in
[4] but with a small change. Zhang [1] observes that it is
convenient to introduce the condition d[T with T being
the product of primes less than x
with = 1/1168.
With this understanding, the terms S
1
and S
2
are easily
handled by direct expansion of the square. Indeed, the
rst term for S
1
is
qD
q|P
rD
r|P
(q)(r)g(q)g(q)
nx
[q,r]|P(n)
1.
Following Zhang, let
1
(q) be the number of solutions of
the congruence P(n) 0 (mod q) for q squarefree and
zero otherwise. By the Chinese remainder theorem, this
is a multiplicative function and we have
1
(p) = k if p
is coprime to
1i<jk
[h
i
h
j
[
and in general,
1
(p) k. Thus, the innermost sum is
x
1
([q, r])
[q, r]
+ O(
1
([q, r])),
where the implied constant is bounded by unity. The
main term of S
1
is
x
qD
q|P
rD
r|P
(q)(r)g(q)g(r)
([q, r])
[q, r]
+ O(D
2+
),
for any > 0 since ([q, r]) k
([q,r])
, where (n) de-
notes the number of distinct prime factors of n. Elemen-
tary number theory shows that this function is O(D
)
for q, r D and any > 0.
GENERAL ARTICLE
We let d
0
= (q, r) and write q = d
0
d
1
, r = d
0
d
2
with
(d
1
, d
2
) = 1. The sum S
1
now becomes
T
1
x + O(D
2+
),
where
T
1
=
d
0
|P
d
1
|P
d
2
|P
(d
1
)(d
2
)
1
(d
0
d
1
d
2
)
d
0
d
1
d
2
g(d
0
d
1
)g(d
0
d
2
) .
The point is that the same sum without the restriction
d
i
[T for i = 0, 1, 2 has already been studied in [4] and
the initial section of Zhangs paper is devoted to show-
ing that (essentially) the same asymptotic formula of [4]
(namely the formula for T
1
given above) still holds with
the extra condition d[T. (More precisely, what is de-
rived is an upper bound for T
1
which is within e
1200
of
T
1
.)
To derive a lower bound for S
2
, we have (after a minor
change of variables)
S
2
=
k
i=1
nx
(n)(n h
i
)
2
.
We expand the square, interchange summation and ob-
tain
k
i=1
q|P
r|P
(q)(r)g(q)g(r)
nx
[q,r]|P(nh
i
)
(n).
To handle the innermost sum, we observe that the con-
dition
P(n h
i
) 0(mod d) (n, d) = 1
is equivalent to n c (mod d) for some c C
i
(d). For
d = p, a prime, this number is the number of distinct
residue classes (mod p) occupied by the set h
i
h
j
:
h
i
, h
j
(mod p) which is
p
(H) 1. Thus, dening
a multiplicative function (supported only on squarefree
GENERAL ARTICLE
values of d)
2
(d) by setting
2
(p) =
p
(H) 1 and
extending it multiplicatively, we obtain (using notation
introduced earlier), the innermost sum as
cC
i
([q,r])
nx
nc( mod [q,r])
(n)
=

2
([q, r])
([q, r])
nx
(n) +
cC
i
([q,r])
(; [q, r], c).
Now it is an elementary exercise (see for example, [2]) to
show that the number of pairs q, r such that [q, r] = d
is given by the divisor function
3
(d) which is the num-
ber of ways of writing d as a product of three positive
integers. Thus, we can simplify this to
nx
(n)(n h
i
)
2
= T
2
nx
(n) + O(E
i
),
where
T
2
=
q|P
r|P
(q)g(q)(r)g(r)
([q, r])

2
([q, r])
and
E
i
=
d<D
2
,d|P
3
(d)
2
(d)
cC
i
(d)
[(, d, c)[.
The error term is estimated using Theorem 2. Indeed,
by the CauchySchwarz inequality
E
i

d<D
2
cC
i
(d)
2
3
(d)
2
2
(d)[(; d, c)[
1/2

d<D
2
,d|P
cC
i
(d
)
[(; d, c)[
1/2
.
GENERAL ARTICLE
On the rst factor, we use the trivial estimate
[(; d, c)[
x
d
+ 1
and see (by elementary number theory) that its contri-
bution is at most a power of a logarithm. For the second
factor, we use Theorem 2 to save the few powers of log-
arithm. Thus,
S
2
= kT
2
x + O(x(log x)
A
).
As before, we can rewrite T
2
as
T
2
=
d
0
|P
d
1
|P
d
2
|P
(d
1
d
2
)
2
(d
0
d
1
d
2
)
(d
0
d
1
d
2
)
g(d
0
d
1
)g(d
0
d
2
).
Again, this sum without the restriction d
i
[T i = 0, 1, 2
was treated in [4] and shown to have the asymptotic
behaviour given by T
2
. Zhang shows that T
2
does not
dier much from T
2
(more precisely, that it is within a
factor of e
1181.579
of T
2
). Thus,
S
2
(log 3x)S
1
Sx(log D)
k+2+1
+o(x(log x)
k+2+1
),
where
=
1
(k + 2)!

2(2 + 1)k
( + 1)(k + 2 + 1)

1
nearly. For = 1/4 + 1/1168, it is easily veried that

> 0.
This calculation already shows that if we have the Elliott
Halberstam conjecture, then for some k
0
and any admis-
sible k
0
-tuple H, the set n + h : h H contains at
least two primes for innitely many values of n. (Tao has
labelled this as DHL[k
0
,2].) Farkas, Pintz and Revesz
[16] made this relationship a bit more precise as follows.
Suppose we have for any A > 0,
d<Q
max
(a,d)=1
[E(x, d, a)[
x
log
A
x
,
GENERAL ARTICLE
with Q = x
. We call this the modied Elliott

Halberstam conjecture EH[]. Let j
n
denote the rst
positive zero of the Bessel function (of the rst kind)
J
n
(x) given by the power series
j=0
(1)
j
2
2j+n
j!(n + j)!
x
2j+n
.
Then we may take any k
0
2 which satises the in-
equality
j
2
k
0
2
k
0
(k
0
1)
< 2
and for this k
0
we have DHL[k
0
,2]. The left-hand side is
greater than 1 and tends to 1 as k
0
tends to innity.
4. Variations of BombieriVinogradov Theorem
and Extensions
Since we are unable to prove EH[] for any > 1/2,
we look for some suitable modication. Based on the
works of Motohashi, Pintz and Zhang, Tao [14] makes
the following conjecture which he labels as MPZ[, ]:
let H be a xed k
0
-tuple (not necessarily admissible)
with k
0
2. Fix w and set W to be the product of the
primes less than w. Let b mod W be a coprime residue
class and put I = (w, x
). Let S
I
be the set of squarefree
numbers all of whose prime factors lie in I. Put
b,W
(; q, a) :=
nx,na( mod q)
nb( mod W)
(n)
1
(q)
nx,nb( mod W)
(n)
and C(q) is the set of zeros (mod q) of the polynomial
P(n). Then the conjecture MPZ[, ] is that
q<x
1
2
+
qS
I
aC(q)
[
b,W
(; q, a)[
x
log
A
x
GENERAL ARTICLE
An important idea in all
proofs of the Bombieri
Vinogradov theorem
is to decompose the
von Mangoldt function
into sums of short
sums.
for any xed A > 0. Zhang proved that MPZ[, ]
holds for any 0 < < 1/1168. Apparently, Zhangs
argument can be extended to show that MPZ[, ] is
true provided
207 + 43 <
1
4
.
The relationship of the MPZ conjecture to the DHL con-
jecture is given by (see [14]) the following result. Let
0 < < 1/4 and 0 < < 1/4 + . Let k
0
2 be an
integer which satises
1 + 4 >
j
2
k
0
2
k
0
(k
0
1)
(1 + ) ,
where
:=
1n<
1+4
2
1
2n
1 + 4
k
0
/2 n
j=1
(1+3k
0
log(1+
1
j
)).
Then MPZ[, ] implies DHL[k
0
, 2]. It is the ne tuning
of this theorem along with other observations (regarding
admissible sets) that have led to the numerical improve-
ments in Zhangs theorem.
Thus, to prove MPZ[, ], we may restrict our moduli
to be in the range (x
1/2
, x
1/2+2
) since the initial range
(1, x
1/2
) can be treated using the classical Bombieri
Vinogradov theorem. Also, it is not dicult to see that
can be replacd by the von Mangoldt function since
the contribution from prime powers (squares and higher)
can be shown to be negligible.
An important idea in all proofs of BombieriVinogradov
theorem is to decompose the von Mangoldt function into
sums of short sums. To be precise, let us dene the
Dirichlet convolution of two arithmetic functions f, g to
be
(f g)(n) :=
d|n
f(d)g(n/d).
GENERAL ARTICLE
Writing the von
Mangoldt function as
a sum of divisor
functions reduces the
study of primes in
arithmetic
progressions to the
study of divisor sums
in arithmetic
progressions.
Let L(n) = log n, 1(n) = 1 for all n and set (n) = 1 if
n = 1 and zero otherwise. Then,
= L and = 1.
If we write f
n
to denote the n-fold Dirichlet convolu-
tion, then
=
10
1
9
L
is a fact utilised by Zhang (in his Lemma 6) to decom-
pose the von Mangoldt function into short sums. Let
x
> (2x)
1/10
and write =
x
+
>x
, where in the
rst term, is restricted to [1, x
] and in the second, to

the range > x
. Clearly
10
>x
1
9
L = 0,
since n x cannot be factored as a product of 10 terms
each larger than x
. Thus, writing
>x
=
x
and
using the binomial formula we see easily that
=
10
j=1
(1)
j1
10
j

j
x
1
(j1)
L
which is an identity of Heath-Brown (but the idea of
decomposing arithmetic functions in this way goes back
to Linnik).
One can also use the formal identity
(s)
(s)
=

(s)
1 + ((s) 1)
=
(s)(1+((s) 1) ((s) 1)
2
+ ).
This allows one to write the von Mangoldt function as a
sum of divisor functions and in this way one reduces the
study of primes in arithmetic progressions to the study
of divisor sums in arithmetic progressions.
In any case, one uses this decomposition of in the
treatment of d in the range [x
1/2
, x
1/2+2
]. The de-
composition leads to three kinds of sums (called Type I,
GENERAL ARTICLE
Suggested Reading
[1] Y Zhang, Bounded gaps between primes, Annals of Mathematics,
2013.
[2] M Ram Murty, Problems in analytic number theory, 2nd Edition,
Springer, 2008.
[3] H Davenport, Multiplicative number theory, 3rd Edition, Springer,
2000.
[4] D Goldston, J Pintz and C Y Yildirim, Primes in Tuples, I, Annals of
Math., Vol.170, pp.819862, 2009.
[5] E Bombieri, J Friedlander and H Iwaniec, Primes in arithmetic
progressions to large moduli, Acta Math., Vol.156, pp.203251, 1986.
Zhangs analysis leads
to the question of
estimating
hyper-Kloosterman
sums for which
Bombieri and Birch
have given
estimates using
Delignes work on the
Weil conjectures.
II and III in the literature, not to be confused with the
types occurring in the Vaughan method). The rst type
involves convolutions , where is supported on the
interval [x
3/8+8
, x
1/24
] which forces the argument of
to be in [x
1/2+4
, x
5/88
]. Type II sums again in-
volve convolutions of the form , but with now
supported on [x
1/24
, x
1/2
] so that is supported on
[x
1/2
, x
1/2+4
], and Type III are the remaining types.
In 1976, Motohashi [17] derived a general induction prin-
ciple to derive theorems of BombieriVinogradov type
for a wide class of arithmetical functions. Much of the
treatment of these types of sums follows earlier work of
Bombieri, Friedlander and Iwaniec [5] and one needs to
verify that the estimates are still valid with the extra
condition d[P. The point to note is that in the range
under consideration, namely d > x
1/2
, the condition
that d[P means we can factor d as d = rq with r lying
in a suitable interval. This factorization turns out to
be crucial in the estimates. Thus, smoothness of d is
essential in this part of the argument.
Another noteworthy point involves Zhangs estimation
of type III sums. His analysis leads to the question of
estimating hyper-Kloosterman sums for which Bombieri
and Birch have given estimates using Delignes work on
the Weil conjectures. Therefore, this work on the twin
prime problem is very much a 21st century theorem!
GENERAL ARTICLE
Address for Correspondence
M Ram Murty
Department of Mathematics
Queens University
Kingston, Ontario
K7L 3NR, Canada
Email:
murty@mast.queensu.ca
progressions to large moduli II, Math. Annalen, Vol.277, pp.361
393, 1987.
progressions to large moduli III, J. Amer. Math. Soc., Vol.2, pp.215
224, 1989.
[8] G H Hardy and J E Littlewood, Some problems of Partitio Numerorum
III: on the expression of a number as a sum of primes, Acta Math.,
Vol.44, pp.170, 1923.
[9] G H Hardy and E M Wright, An introduction to the theory of numbers,
6th Edition, Oxford University Press, 2008.
[10] H G Gadiyar and R Padma, Ramanujan-Fourier series, the Wiener-
Khintchine formula and the distribution of prime pairs, Phys. A.,
Vol.269, No.24, pp.503510, 1999.
[11] B Bagchi, A promising approach to the twin prime problem, Reso-
nance, Vol.8, No.3, pp.2631, 2003.
[12] P D T A Elliott and H Halberstam, A conjecture in number theory,
Symp. Math., Vol.4, pp.5972, 1968.
[13] P Dusart, Autour de la fonction qui compte le nombre de nombres
premiers, PhD Thesis, Universit de Limoges, 1998.
[14] T Taos blog:terrytao.wordpress.com/tag/yoichi-motohashi/
[15] michaelnielsen.org/polymath1/index.php?title=Bounded\_gaps\_
between\_ primes
[16] B Farkas, J Pintz and S Revesz, On the optimal weight function in the
GoldstonPintzYildirim method for finding small gaps between
consecutive primes, to appear.
[17] Y Motohashi, An induction principle for the generalization of
Bombieris prime number theorem, Proc. Japan Acad., Vol.52, No.6,
pp.273275, 1976.

Twin Prime No

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Twin Prime No

Enviado por

Direitos autorais:

Formatos disponíveis

712 RESONANCE August 2013

which resembles 1/(s + 1) (with the 2-Euler factor re-

(0) and because (s + 1) has a simple pole at

contains at least two primes. Consequently,

. The term in brackets is positive

with = 1/4 + 1/1168.

nearly. For = 1/4 + 1/1168, it is easily veried that

. We call this the modied Elliott

] and in the second, to

Você também pode gostar